The KV-cache is the right memory for datacenters but the wrong memory for robots. Datacenter inference batches many short requests and resets them, amortizing an attention cache across a crowd. Embodied agents instead run one long, non-resetting episode on bandwidth-limited edge hardware, where high-bandwidth memory and flash are scarce, flash has finite write endurance, and memory writes rather than compute can become the binding constraint. AURA-Mem (Action-Utility Recurrent Adaptive Memory) targets this regime. It wraps a frozen vision-language-action backbone with a constant-size recurrent memory and a learned gate that writes only when the current observation would change the next action: memory that knows when to stay silent. Unlike reconstruction-based memory, the gate is trained directly against a closed-loop action-error signal. Its inference state is fixed at 4,224 bytes regardless of horizon, while a KV-cache grows to 6,061 times larger at 100,000 steps. On a controlled synthetic benchmark, AURA-Mem matches the best O(1) baseline in accuracy while using 5.19-6.13 times fewer writes, and up to 9.19 times fewer writes on easier configurations. Budget-matched random and periodic schedules do not recover this gain, isolating the benefit to the action-surprise signal. On a trained closed-loop OpenVLA-OFT 7B panel on LIBERO-Long (n=60 episodes per arm), the gate does not hurt success: AURA-Mem matches the ungated base policy (0.233) and slightly exceeds an always-write KV arm (0.217), while using 7.0 times fewer writes and constant memory. We also instantiate an approximate-information-state value-loss bound as a methodology demonstration; at this scale, the bound is vacuous rather than a guarantee.

翻译：KV缓存是数据中心适用的记忆，却并非机器人适用的记忆。数据中心推理时批量处理大量短请求并重置它们，在群体间分摊注意力缓存。而具身智能体则在带宽受限的边缘硬件上运行一个连续不断且不复位的长线程，此处高带宽内存和闪存稀缺，闪存写入寿命有限，内存写入而非计算可能成为约束瓶颈。AURA-Mem（动作效用循环自适应记忆）针对这一场景设计。它用一个恒定大小的循环记忆和一个学习到的门控机制来包裹冻结的视觉-语言-动作骨干网络，该门控仅在当前观测会改变下一动作时执行写入：一种懂得何时保持静默的记忆。与基于重建的记忆不同，该门控直接针对闭环动作误差信号进行训练。其推理状态固定为4,224字节，不受时间跨度影响，而KV缓存则在10万步时增长至其6,061倍。在受控的合成基准测试中，AURA-Mem在精度上匹配最佳O(1)基线，同时写入次数减少5.19至6.13倍，在较简单配置下写入次数最多减少9.19倍。预算匹配的随机和周期性调度无法恢复这一增益，从而将效益归因于动作意外信号。在LIBERO-Long数据集上训练的闭环OpenVLA-OFT 7B面板测试中（每臂n=60个线程），该门控并未损害成功率：AURA-Mem匹配无门控基础策略（0.233），并略优于始终写入的KV方案（0.217），同时写入次数减少7.0倍且内存恒定。我们还实例化了一个近似信息状态值损失界限作为方法论演示；在此规模下，该界限是空洞的而非保证。